Re: Uninline kcalloc

[Xi Wang]

On Feb 16, 2012, at 9:51 AM, Christoph Lameter wrote:
> Then there is
> 
> vzalloc
> kzalloc
> vmalloc_32
> alloc_bootmem (MAXORDER limit may not work)
> alloc_remap
> 
> ...
> 
> This would also work for special subsystem allocations like
> 
> usb_alloc_coherent
> dm_vcalloc
> devres_alloc
> 
> ....
> 
> 
> The use of a function or macro makes the overflow check much more
> universal and allows these array allocations to occur with different
> allocation functions throughout the kernel.

No, it does NOT.  It can be easily misued to introduce more bugs.

1) Should calculate_array_size() return 0 on overflow, as you
   orginally proposed?

No, as Dan, Pekka, and some others already pointed out.

2) Should calculate_array_size() return something like
   KMALLOC_MAX_SIZE + 1?

No, because you intended to use it with other allocators such as
vmalloc().

3) Should calculate_array_size() return ULONG_MAX/SIZE_MAX/-1?

No!  Consider devres_alloc() you mentioned.  Then you do

devres_alloc(..., calculate_array_size(n, size), ...).

It internally invokes kmalloc() with allocation size:

sizeof(struct devres) + calculate_array_size(n, size).

When n * size overflows, calculate_array_size() returns ULONG_MAX,
and the allocation size wraps around to a small integer!

I like the idea of "do not add an allocator unless necessary".
However, "universal" calculate_array_size() just doesn't work,
unless you can find the correct semantics or limit its use.
It can be easily misused and bring more trouble than it's worth.

- xi

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/